JP2014063372A - Apparatus and method for predicting power generation capacity - Google Patents

Abstract

PROBLEM TO BE SOLVED: To accurately predict power generation capacity of PV systems.SOLUTION: According to an embodiment of the present invention, a power generation capacity prediction apparatus includes; a weather condition prediction unit which derives a predicted weather condition value for each PV system by predicting weather condition at each PV system installation site from a system coefficient and power generation history data; a weather condition spatial correction unit which obtains a corrected weather condition value for each PV system by correcting the predicted weather condition value in accordance with an installed position of each PV system; a parameter learning unit which updates the system coefficient of each PV system based on the corrected weather condition value and the power generation history data; and a power generation prediction unit which predicts power generation capacity of each PV system based on reference weather data and the system coefficient.

Description

  Embodiments described herein relate generally to a power generation amount prediction apparatus and method.

  Photovoltaic (PV) system is a power generation facility, and power generation revenue is determined according to the amount of power generation and the purchase price. At this time, it is necessary to make a long-term prediction of power generation over several decades considering the life cycle of the PV system. Given the long-term purchase price of electricity, if the PV system's power generation capacity and the weather environment of the site where the PV system exists can be modeled, it will be possible to analyze the amount of power generation and generation revenue. Conventionally, a method for predicting the amount of power generation by modeling the power generation capacity of a PV system from the power generation result data has been proposed.

D. King, W. Boyson, and J. Kratochvil, Photovoltaic Array Performance Model, SAND2004-3535, Sandia National Laboratories, Albuquerque, NM, December 2004.

  It is important to analyze the long-term power generation outlook of the PV system using the actual power generation data, but the conventional method uses a simple linear regression model to correct the system parameters only from the actual power generation data of a single PV system. It was difficult to eliminate the effects of noise from sensors and shadows from buildings. For this reason, there has been a problem that it is impossible to perform long-term prediction of power generation with high accuracy.

  One aspect of the present invention has been made to solve the above-described problems, and an object thereof is to provide a power generation amount prediction apparatus and a method thereof capable of performing highly accurate power generation amount prediction of a PV system. There is to do.

  A power generation amount prediction apparatus as one aspect of the present invention includes a reference weather data storage unit, a power generation result data storage unit, a PV system coefficient storage unit, a weather condition estimation unit, a weather condition space correction unit, and a power generation prediction unit And comprising.

  The reference weather data storage unit stores reference weather data including weather situation values collected at a reference weather site.

  The power generation result data storage unit includes a current value and a voltage value of power generated by a photovoltaic power generation (PV: Photovoltaic) system installed at a plurality of sites different from the reference weather site. Store the data.

  The PV system coefficient storage unit stores a system coefficient that is a coefficient for predicting a power generation amount from a weather condition value for each of the PV systems.

  The weather condition estimation unit obtains an estimated weather condition value for each of the PV systems by estimating the weather condition at the site from the system coefficient and the power generation record data.

  The weather condition space correction unit obtains a corrected weather condition value by performing a correction process on the estimated weather condition value according to an installation position of each PV system.

  The parameter learning unit updates the system coefficient for each PV system based on the corrected weather condition value and the power generation record data.

  The power generation prediction unit performs power generation amount prediction for each PV system based on the reference weather data and the system coefficient.

The block diagram of the whole system provided with the PV electric power generation long-term prediction apparatus which concerns on one Embodiment of this invention. The block diagram of the PV electric power generation long-term prediction apparatus which concerns on one Embodiment of this invention. The block diagram of the computer system which concerns on one Embodiment of this invention. The flowchart of the operation | movement concerning one Embodiment of this invention. The figure which shows the example of a PV system coefficient. The figure which shows IV characteristic curve. The figure which shows the example of reference weather data. The figure which shows the table | surface of power generation performance data. The figure which shows the graph of the electric power generation performance data of several PV site. The figure which shows the example of the area weather environmental coefficient regarding the amount of solar radiation (sunshine value) and temperature. The figure which shows the example of the map of a solar position weather correction coefficient. The figure which shows the electric power generation amount prediction result about a certain PV site. The figure for demonstrating weather condition space correction | amendment. Explanatory drawing of the selection method of power generation performance data.

  Embodiments of the present invention relate to a long-term power generation amount prediction device for a photovoltaic power generation system (hereinafter referred to as a PV system), and in particular, a power generation result data of a plurality of PV systems after starting operation and a reference weather site such as a weather station By using long-term meteorological data, it is possible to eliminate the effects of noise from PV site meteorological sensors and shadows on PV systems caused by buildings, etc. It relates to a long-term prediction device.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  First, the basic matters of long-term power generation forecast are explained.

  The power generation capacity of the PV system installed at site i can be represented by a performance model using system parameters θi. The weather environment at site i can be represented by a probability model Ei. Then, for example, the power generation amount of this PV system from the current time T to the time point Te can be predicted by the function predict (θi, Ei | T˜Te).

When designing a PV system, the weather environment model Ei of the site to be installed is not obtained. For this reason, the weather environment model Er at the reference weather site is generated using the weather data er of the reference weather site such as a nearby weather station. And using the system parameter θo representing the power generation capacity of a standard PV system,
predict (θo, Er | T〜Te) (1)
The power generation amount prediction in the time range T to Te can be performed.

  When the PV system actually starts operating at site i, power generation result data di is obtained. The power generation result data d can be represented by a set of data {d (t)} at each time t. The data at each time is d (t) = {I (t), where the generated DC current I (t), the generated DC voltage V (t), the sunshine value S (t), and the temperature C (t) are combined. V (t), S (t), C (t)} etc. Moreover, only d (t) = {I (t), V (t)} may be obtained in a small-scale PV site. The collected time range varies depending on the site. If the current time is T, the power generation result data is accumulated in the range of (Ti ≦ t ≦ T).

Estimate system parameter θi using power generation data consisting of di (t) = {I (t), V (t), S (t), C (t)}
predict (θi, Er | T〜Te) (2)
The power generation amount prediction reflecting the power generation result data can be performed.

  However, when making such a prediction, it is possible to make an accurate prediction if the assumption that the weather sensor data at the site where the PV system is installed is correct and there is no influence of the shadow of the neighborhood is not satisfied. There is a problem that it is not possible. The embodiment of the present invention solves these problems by combining the power generation result data and reference weather data of a plurality of sites.

  FIG. 1 shows a system example of a long-term power generation amount prediction device for a solar power generation system assumed by an embodiment of the present invention.

  In FIG. 1, a reference weather site 202 represents a weather station and the like, and long-term weather data {er (t)} is collected. The weather data can be considered as a sample from the assumed weather condition model Er. Usually, the reference weather site is on a hill or the like and is not easily affected by shadows. Also, meteorological observatory sensors (such as a pyranometer, thermometer, etc.) are usually more accurate than PV system meteorological sensors.

  In the weather target area 205, a plurality of PV systems to be managed are installed at each site. In the target area 205 of FIG. 1, four PV systems including the PV system 203 are managed. Each PV system is assigned power generation result data d, system parameter θ, solar position weather correction parameter M, and area weather environment parameter E. The common area weather environment parameter Ea may be used as the area weather environment parameter of each PV system. Details of each parameter will be described later. Accumulation of these data and parameters, or calculation of parameters is performed by the calculation center 204. This represents a calculation service providing facility such as a cloud. The PV power generation long-term prediction device according to the embodiment of the present invention can be realized as a program module on the calculation center 204.

  FIG. 2 is a block diagram showing an embodiment of a PV power generation long-term prediction apparatus according to an embodiment of the present invention. As shown in FIG. 2, the power generation amount prediction apparatus includes a reference weather data storage unit 301, a weather condition space correction unit 302, a weather condition estimation unit 303, a power generation result data selection unit 304, a solar position calculation unit 305, and a power generation result. A data storage unit 306, a parameter learning unit 307, a solar position weather correction coefficient storage unit 308, an area weather environment coefficient storage unit 309, a PV system coefficient storage unit 310, and a power generation prediction unit 311 are provided.

  Each component of the apparatus in FIG. 2 can be realized as, for example, a program module. In this case, the function can be realized by executing a program including each program module in the computer system shown in FIG. The computer system includes a CPU 101 that executes program instructions, a main storage device 104 such as a memory, an external storage device 103 such as a hard disk, a magnetic disk device, or a magneto-optical disk device, an input device 105 that inputs data by a user, A display device 106 for displaying data and a bus 102 for connecting them to each other are provided. The program is stored in the external storage device 103, and the CPU 101 expands the program in the main storage device 104, and sequentially reads and executes the expanded program.

  The PV system coefficient storage unit 310 in FIG. 2 stores the PV system power generation amount p (t) = {I (t), V (t from the weather condition e (t) = {S (t), C (t)}. )} Is stored for predicting parameters (system coefficients). FIG. 5 shows an example of the PV system coefficient. In FIG. 5, 502 represents a PV system coefficient for a site where a weather sensor exists, and 503 represents an example of a PV system coefficient for a site where a weather sensor does not exist.

I _ph , I ₀ , α, and Rs are diode circuit coefficients, and Ccoef_V and Ccoef_I are power generation panel temperature coefficients. I _ph is an electromotive force of power generation, I ₀ is a characteristic parameter called reverse saturation current parameter, α is a predetermined coefficient, and Rs is a series resistance of the power generation module. I _ph , I ₀ , α, and Rs can be determined by learning. Alternatively, some or all of these may be given in advance, such as using values in a spec sheet. Ccoef_V and Ccoef_I can be determined by learning. Alternatively, Ccoef_V and Ccoef_I may be given in advance, such as using the values in the spec sheet. C0, C1, S0, and S1 are weather sensor correction coefficients, which are determined by learning. The parameter learning method will be described later.

  An example of a procedure for predicting the power generation amount using the PV system coefficient 502 as an example is shown below.

によって、IVカーブを算出する。これは、準ニュートン法などの反復計算を用いて行うことができる。図６にIVカーブの例６０１を示す。図６の６０２はX軸に電圧(V)をY軸に電流(I)をとったIVカーブであり、PVシステムの発電能力を表す。 Step A1) Using parameters (I _ph , I ₀ , Rs, α)

To calculate the IV curve. This can be done using iterative calculations such as the quasi-Newton method. FIG. 6 shows an example IV curve 601. 602 in FIG. 6 is an IV curve in which the voltage (V) is taken on the X axis and the current (I) is taken on the Y axis, and represents the power generation capacity of the PV system.

Step A2) Obtain the maximum power point (Im, Vm) of the IV curve. In FIG. 6, 603 corresponds to the maximum power point. The PV system shall perform maximum power point control (MPPT control) with a power conditioner in order to maximize the generated power.

Step A3) The weather sensor value is corrected as follows.

C '(t) = (C (t)-C0) / C1, S' (t) = (S (t)-S0) / S1 (4)

Step A4) The amount of power generation is predicted as follows. Note that “25” in Equation (5) and Equation (6) represents the reference temperature of the power generation module.

I ^* (t) = Im * S '(t) / (1 + Ccoef_I * (C' (t)-25)) (5)
V ^* (t) = Vm / (1 + Ccoef_V * (C '(t) -25)) (6)

  The upper step A3 is a correction process assuming that there is a problem with the weather sensor value. This is important because some weather sensors can have problems with accuracy.

  By repeating such processing by changing the time t, the power generation amount p = (p (p ( 0), ..., p (t), ...) can be predicted. Further, at the PV site where no weather sensor exists, the above step A3 may be omitted.

  The reference weather data storage unit 301 in FIG. 2 stores long-term accurate weather data such as a weather station in the vicinity of the target area. Reference weather data includes sunshine values and temperatures. Other types of weather condition values may be included.

  FIG. 7 shows an example 901 of reference weather data, where 902 corresponds to the air temperature and 903 corresponds to the sunshine value. In general, it can be expected that the meteorological data at reference weather sites such as weather stations is abundant. The example in FIG. 7 includes data from January 1, 1980. However, the data on the reference weather site may be different from the environment where the PV site is installed, including the influence of shadows, etc. Therefore, when calculating the weather conditions of PV sites, it is necessary to make corrections that take into account the differences. In that case, it is important to use the power generation result data of a plurality of sites.

  The power generation record data storage unit 306 in FIG. 2 stores the power generation record of the PV system in the target area. FIG. 8 shows an example 1001 of power generation result data of a certain PV site (PV system) in a tabular form. The power generation result data storage unit 306 stores power generation result data as shown in FIG. 8 for each PV site.

  In FIG. 8, the power generation results at a certain point in each row are shown, and the data are arranged in time order. The data consists of fields of DC voltage, DC current, sunshine value, temperature, corrected sunshine value, corrected air temperature, and weight.

  In the DC voltage field and the DC current field, values of DC voltage and DC current relating to the generated power are stored. The values of the DC voltage field and DC current field need to be included in all PV sites.

  Also, at PV sites with weather sensors, sensor values are stored in the sunshine value field and temperature field. For PV sites without a weather sensor, the estimated value calculated by the method described later (formulas D1 to D4) is stored.

  The corrected sunshine value field and the corrected temperature field store values obtained by correcting the values of the sunshine value and the temperature field as described later.

  Furthermore, the weight field stores a weighting coefficient that represents the importance of data at the time of parameter learning. If the value is 0, the data may not be used for learning. As will be described later, a value of 0 is set for data having a large influence of shadow. The weighting coefficient is stored during parameter learning.

  FIG. 9 shows the sunshine value and the temperature in a graph format among the power generation result data of a plurality of PV systems. It can be seen that the collection period varies depending on the start date of each PV system.

  In the area weather environment coefficient storage unit 309 in FIG. 2, a weather probability model of the target area (see 205 in FIG. 1) is accumulated. In other words, weather condition data e as shown in FIG. 7 or FIG. 9 can be sampled from this model. In addition, by introducing a weather probability model, it becomes possible to obtain a probability distribution of the amount of power generation, as will be described later.

  FIG. 10 shows an example 1201 of the area weather environment coefficient. Here, at each time of the year and month of 365, one day, the coefficient relating to the sunshine value and the temperature is accumulated. These coefficients represent normal distribution parameters. μr represents the average of the reference weather site, and σr represents the standard deviation of the reference weather site. Further, μa represents the average of the difference in weather conditions from the reference weather site for the target area. For example, the value of (250, 18, 20) is accumulated for the sunshine value at 8:00 on 1/1. That is, if the normal distribution of mean μ and standard deviation σ is expressed as N (μ, σ), the probability distribution of the sunshine value at the relevant time at the reference weather site (meteorological station etc.) is N (250, 18) The area represents that the sunshine value (sunshine amount) can be sampled by a normal distribution of N (250 + 20, 18) = N (270, 18). The temperature can be sampled in the same way. In the example of FIG. 10, sunshine is expressed in a normal distribution every hour, but may be every day or every 3 hours. A distribution other than the normal distribution may be assumed. A method for calculating μr, σr, and μa will be described later.

  The weather condition of the target area is basically represented by a probability model expressed by such area weather environment parameters. However, the assumption that the individual weather environment of each PV site is all equal within the target area may not hold in urban areas. The reason for this is the influence of shadows from surrounding objects. This is because the amount of sunlight reaching the PV system is reduced by the surrounding shadow. Shadows also lower the PV site temperature. The effects of surrounding shadows are not temporary and must be reflected in the long-term forecast of power generation.

  In the embodiment of the present invention, only the sunshine value and the temperature of the surrounding objects are considered with respect to the influence on the weather condition of each PV site. Then, it is considered that the influence should be modeled for each solar position.

  In the solar position meteorological correction coefficient storage unit 308 in FIG. 2, correction parameters (correction coefficients) are accumulated for each PV site for each range divided by a grid for each solar azimuth and altitude.

  FIG. 11 shows an example 701 of the solar position weather correction coefficient. A map in which a coefficient of 0 to 1 is set for each grid is shown in the range divided by the grid for each direction and altitude of the sun. Here, a coefficient 0.3 regarding sunlight and a coefficient 0.95 regarding temperature are stored in the upper right grid. Each of the other grids stores a coefficient relating to sunlight and a coefficient relating to temperature. The coefficient value is determined by an estimation method described later. Since the position of the sun is determined according to the date and time, a map in which a coefficient of 0 to 1 is set according to the date and time can be prepared and used.

  Here, regarding the sunshine value of the target area at time t, the area weather environment coefficient is represented by {μr (t), σr (t), μa (t)}. Further, it is assumed that the solar position sunshine correction coefficient of the target site is represented by M (φ) using the sun position φ. At this time, the sunlight value can be sampled as follows using this solar position weather correction coefficient. The temperature can be similarly sampled.

Step B1) e ′ (t) is sampled by e ′ (t) to N (μr (t) + μa (t), σr (t)). That is, data is sampled from a normal distribution of N (μr (t) + μa (t), σr (t)).

Step B2) Calculate φ by φ = sun_dir (t). However, sun_dir (t) is a function for calculating the sun position at time t of the target site.

Step B3) e ′ (t) is corrected by e (t) = M (φ) * e ′ (t).

By performing such processing on the sunshine value and the temperature, it is possible to obtain a random sample of weather data in consideration of the influence of the shadow of the surrounding object. Then, the prediction result of the electric power generation amount in each sample can be obtained by performing the processing of steps A1) to A4) described above for each of the samples.
The power generation prediction unit 311 in FIG. 2 performs long-term power generation prediction using such a prediction result. FIG. 12 shows an example 1401 of the power generation amount prediction result for a certain PV site. In this example, the power generation amount for 6 years is predicted.

  In FIG. 12, 1402 represents the power generation amount for a single year, and 1403 represents the cumulative power generation amount. The center solid line represents the center point (average), and the dotted line represents the 95% confidence interval. In the graph of FIG. 12, the accumulated power generation starts from below the single-year power generation because the scales of the vertical axes are different.

  Such section estimation of the power generation amount is possible by using a plurality of samples. That is, weather conditions (sunshine value, temperature) are sampled multiple times at the same time from the weather environment model of the target area to obtain multiple samples, the samples are corrected according to the sun position, and the formula from the corrected sample Generate multiple power generation samples for the same time by repeating the process of calculating the IV curve using (1), finding the maximum power point, and predicting the power generation amount using Equation (5) and Equation (6) To do. Average and confidence intervals are determined from these power generation samples.

  Note that the sun position calculation unit 305 in FIG. 2 has a function of calculating the sun position according to time t. The power generation prediction unit 311 uses the sun position calculation unit 305 to calculate the sun position according to time t. That is, step B2 described above is executed.

These power generation amount prediction processes use the system parameter θi of PV site i and the weather model Ei of PV site i,

predict (θi, Ei | T〜Te) (7)
This is equivalent to making a prediction. That is, from time T to time Te, the power generation amount of the PV site i is predicted based on the system parameter θi of the PV site i and the weather model Ei of the PV site i. By using the model unique to the site i, it is possible to perform prediction with higher accuracy than the above-described equations (1) and (2). However, the prediction using Expression (2) can be predicted with higher accuracy than Expression (1), and is included in the prediction of this embodiment.

  As described above, it is possible to perform power generation prediction with high accuracy by using the solar position weather correction coefficient 308, the area weather environment coefficient 309, and the PV system coefficient 310.

  These coefficients can be determined by performing learning in the parameter learning unit 307 using the data in the power generation result data storage unit 306 in FIG. However, for that purpose, various data necessary for parameter learning must be added to the actual power generation data collected at the PV site (DC current, DC voltage, sensor value if there is a weather sensor). What is particularly needed is an accurate estimate of the weather conditions (sunshine value, temperature) at each point in time for each PV site.

  The weather condition estimation unit 303 in FIG. 2 estimates the weather condition for each PV site using the power generation result data and the PV system coefficient. There are cases where meteorological sensors are installed and not installed at PV sites. Examples of estimation methods are described for each case.

<When a weather sensor is installed>

C1) C ^ (t) = (C (t)-C0) / C1
C2) S ^ (t) = (S (t)-S0) / S1

Thus, it is possible to estimate the estimated temperature C ^ (t) and the estimated sunshine S ^ (t).

<When no weather sensor is installed>

D1) Using the parameters (I _ph , I ₀ , Rs, α), calculate the IV curve (the above formula (3)).
D2) Find the maximum power point (Im, Vm) of the IV curve.
D3) C ^ (t) = (Vm / V (t)-1) / Ccoef_V + 25
D4) S ^ (t) = (I (t) / Im) * (1 + Ccoef_I * (C ^ (t)-25))

Thus, it is possible to estimate the estimated temperature C ^ (t) and the estimated sunshine S ^ (t).

  In this way, it is possible to estimate the weather conditions at multiple PV sites. These estimation results are stored in the corresponding fields of the power generation performance data ("sunshine" and "temperature" fields in FIG. 8). When a weather sensor exists at the PV site, the initial sensor value is stored in the field, but this sensor value may be overwritten by the estimated value. Alternatively, another field (not shown in FIG. 8) may be added to store the estimated value therein.

  The weather condition space correction unit 302 in FIG. 2 spatially arranges the weather condition data (sunshine value and temperature) estimated by the weather condition estimation unit 303 using the site position information, and the true value is obtained by the space correction process. Is estimated. Here, it is assumed that the weather conditions at sites that are sufficiently close to each other are sufficiently close if the influence of local shadows or the like is excluded, and the true value is estimated.

  FIG. 13 shows an example 801 of weather condition space correction. Each numerical value in the figure represents the weather condition value (sunshine value or temperature) of each PV site. In this example, correction processing is performed by a simple spatial average. As other spatial correction processing, spatial moving average, spatial median, smoothing by Markov random field, etc. can be applied. In a simple spatial average, the correction value of each PV site is the same. However, in other methods such as a spatial moving average, it is a matter of course that the correction value may differ for each PV site.

  When there is no weather sensor, or due to the influence of the noise of the weather sensor, the estimation of the weather condition may be inaccurate only by the weather condition estimation value of each PV site. Thus, by performing such a spatial correction process, it is possible to estimate the weather condition with high accuracy while eliminating the adverse effects of sensor noise. These corrected weather condition estimated values are stored in the corresponding fields (fields of “corrected sunshine” and “corrected air temperature” in FIG. 8) of the power generation result data.

  The spatial correction described above assumes the spatial continuity of weather conditions. However, in some cases, the weather conditions at one site may deviate significantly from others due to local shadows and the like. Since parameter estimation using such outliers has an adverse effect, it is necessary to select data used for parameter estimation (estimation of system coefficients, area weather environment coefficients, etc.). In order to consider only the influence of the local weather environment, the solar position weather correction factor is used for selection.

  The power generation result data selection unit 304 in FIG. 2 uses the solar position calculated by the solar position calculation unit 305 to calculate the weight coefficient so that the importance of the data increases as the value of the solar position weather correction coefficient is closer to 1. To do. For example, the weight coefficient is set to 1 for data whose solar position weather correction coefficient is equal to or greater than the threshold, and the weight coefficient is set to 0 for data less than the threshold.

  FIG. 14 shows an example 1301 of the method for selecting the power generation result data. Depending on the value of the solar position meteorological correction coefficient M (φ), the data in the range of the sun position 1303 (that is, the data in the date and time range of 1303) is 0 and the data in the range of the sun position 1302 (That is, data in a date and time range of 1302) has the weighting factor set to 1. The weighting coefficient set by such a method is stored in a corresponding field (“weight” field in FIG. 8) of the power generation result data.

  The parameter learning unit 307 in FIG. 2 learns parameters for each PV site using the power generation result data stored in the power generation result data storage unit 306. The parameter learning unit determines the solar position sunshine correction coefficient M (φ), the area weather environment coefficient Ea (μr, σr, μa), and the PV system coefficient θ.

[Learning Solar Position Sunlight Correction Factor]
First, in learning the solar position sunshine correction coefficient, for example, the weather condition value after spatial correction is regarded as a true value, and the weather condition value after spatial correction and the estimated value before spatial correction (formula C1, C2, formula D1 The parameter can be determined so that the difference from the weather condition value locally estimated by D4 etc. is minimized. For example, if the sun position at time t is φ (t), the estimated sunshine value S ^ (t) = 2.0, and the corrected estimated value S ^* (t) = 12.6,

M (φ (t)) = 2 / 12.6 ≒ 0.16 (8)
Etc. and determine the parameters. Actually, more data (for example, data at the same time on each year, month, day) can be obtained, and the value may be determined by averaging the values obtained from the equation (8). Further, when the number of data is small, Bayesian estimation or the like may be used.

[Learn PV system coefficient θ]
Next, in learning of the PV system coefficient θ, the meteorological sensor correction coefficients (S0, S1, C0, C1 used in Equation (4)) and IV curve parameters are determined.

In the IV curve parameter learning, the maximum power point is (I (t), V (t)) based on the spatially corrected weather estimates (S ^* (t) and C ^* (t)). Then, the parameters (I _ph , I ₀ , Rs, α) may be determined. As a determination method, the least square function may be minimized by a quasi-Newton method, a hill climbing method, or the like. As for (I _ph , I ₀ , α), it is also possible to learn only Rs using the PV panel spec sheet as it is. Note that Ccoef_I and Ccoef_V are given in advance.

Alternatively, the predicted current I ^* (t) and the predicted voltage V ^* (t) (see Equation (5) and Equation (6)) are (I (t), V (t)) so that the minimum The parameters (I _ph , I ₀ , Rs, α, Ccoef_I, Ccoef_V) may be determined so that the square function is minimized by a quasi-Newton method or a hill-climbing method. Some parameters may be used as they are in the spec sheet.

Regarding learning of the weather sensor correction factor, the value obtained by converting the sensor value (S (t), C (t)) using equation (4) is closest to (S ^* (t), C ^* (t)) (S0, S1, C0, C1) may be determined by the least square method.

[Learning area weather environment coefficient]
Finally, in learning of area weather environment coefficients, it is possible to obtain parameters by estimating a probability distribution from a collection of weather data at time t for a plurality of years. At that time, the variation of the data may be estimated from abundant data of the reference weather site.

を時刻tにおいてデータが収集されている当該エリアのPVサイトの個数N(t)で除算することによって求めることが可能である。ここでSr(t)は参照気象サイトの日照値、S^i(t)は各PVサイトでの日照推定値である。あるいは、複数年度分の時刻tのデータが存在する場合は、 (9)式のNによる除算値の確率分布を推定し、その確率分布の平均をμaとしてもよい。気温についても同様にしてパラメータを求めることが可能である。 For example, by collecting the amount of sunshine at t = (1/1 8:00) for the reference weather site for multiple years and estimating the probability distribution from those data, Er (1/1 8:00) = N (μr, σr) can be obtained. And for the difference in sunshine average μa between the reference weather site and the target area at the same time,

Can be obtained by dividing by the number N (t) of PV sites in the area where data is collected at time t. Here, Sr (t) is the sunshine value at the reference weather site, and S ^ i (t) is the estimated sunshine value at each PV site. Alternatively, when there are data at time t for a plurality of years, the probability distribution of the division value by N in equation (9) may be estimated, and the average of the probability distribution may be μa. The parameter can be obtained in the same manner for the temperature.

  In determining these parameters, it is effective to make a difference in the importance of learning data. At that time, by weighting the data according to the weighting coefficient calculated by the power generation result data selection unit 304, it is possible to perform accurate parameter estimation that excludes abnormal values. For example, parameter estimation is performed without using data having a weight of zero.

  FIG. 4 is a flowchart showing an embodiment of the PV power generation long-term prediction system according to the embodiment of the present invention. Note that various parameters (coefficients) to be learned are initially set with initial values such as spec sheet values as needed.

  When the power generation long-term prediction process starts, in step 401, the reference weather data and the power generation result data are read from the reference weather data storage unit 301 and the power generation result data storage unit 306 into the memory.

  In step 402, the power generation result data selection unit 304 selects the power generation result data based on the value of the solar position sunshine correction coefficient related to the sun position. For example, the weight coefficient 1 or 0 is set in the power generation result data by comparison with a threshold value. This process can eliminate the influence of shadows.

  After that, in step 403, the weather condition estimation unit 303 estimates the weather condition (sunshine value, temperature) using the PV system coefficient and the selected power generation result data. When there is a weather sensor at the PV site, estimation is performed using C (t) and S (t) data, and when there is no weather sensor, V (t) and I (t) data are used (Equation (4), (See formulas D3 and D4)

  The above steps 401 to 403 are performed for a plurality of sites. In step 404, the weather condition space correction unit 302 performs space correction (smoothing process) on the weather condition estimated for each site. Thereby, sensor noise can be eliminated.

  Then, using the weather situation data estimated in step 403, the weather situation data corrected in step 404, and the power generation result data, in step 405, the parameter learning unit 307 performs a PV system coefficient, an area weather environment coefficient, And the solar position sunshine correction coefficient is updated. The update method is as described above.

  Since there is a possibility that the results of steps 402 and 403 are different before and after the update due to this update, the parameter learning unit 307 determines in step 406 whether the update amount of the parameter (various coefficients) is sufficiently small. If there is still room for updating, the process returns to step 402 to repeat the process. For example, it is determined that all parameters have converged when the difference from the previous value is less than ε. ε may be set for each parameter, or a common value may be used. Alternatively, it may be determined that only certain parameters have converged, not all parameters, and have converged.

  If the parameters have converged, in step 407, the power generation prediction unit 311 performs power generation prediction for each PV system. The predicted power generation amount is output to a display or the like.

  As described above, according to the embodiment of the present invention, the parameter estimation is performed using the power generation result data and the obstacle model (solar position weather correction coefficient) of a plurality of sites. Long-term prediction of power generation with high accuracy can be performed for the system.

  Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of components disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different embodiments may be appropriately combined.

Claims (11)

A reference weather data storage unit for storing reference weather data including weather condition values collected at the reference weather site;
Power generation result data storage for storing the power generation result data for each PV system, including the current value and voltage value of the power generated by the photovoltaic power generation (PV: Photovoltaic) system installed at a plurality of sites different from the reference weather site And
For each of the PV systems, a PV system coefficient storage unit that stores a system coefficient that is a coefficient for predicting a power generation amount from a weather condition value;
For each of the PV systems, by estimating the weather conditions at the site from the system coefficient and the power generation performance data, a weather condition estimation unit that obtains an estimated weather condition value;
A weather condition space correction unit that obtains a corrected weather condition value by correcting the estimated weather condition value according to the installation position of each PV system;
For each PV system, based on the corrected weather condition value and the power generation performance data, a parameter learning unit that updates the system coefficient;
For each PV system, a power generation prediction unit that performs power generation amount prediction based on the reference weather data and the system coefficient,
A power generation amount prediction device equipped with An area weather environment coefficient storage unit for storing an area weather environment coefficient representing a difference in weather conditions between the reference weather site and a target site including the plurality of sites;
The parameter learning unit updates the area weather environment coefficient based on the difference between the estimated weather condition value and the weather condition value of the reference weather data,
The power generation amount prediction apparatus according to claim 1, wherein the power generation amount prediction unit further performs the power generation amount prediction using the area weather environment coefficient. The power generation amount prediction unit samples the weather condition value of the target area based on the reference weather data and the area environment coefficient, and performs the power generation amount prediction based on the sampled weather condition value and the system coefficient. The power generation amount prediction apparatus according to 2. For each PV system, a correction coefficient storage unit that stores a sunshine correction coefficient associated with the date and time,
The weather condition value includes a sunshine value,
The parameter learning unit updates the sunlight correction coefficient for each PV system so as to minimize a difference between the corrected weather condition value corrected by the sunlight correction coefficient and the estimated weather condition value. ,
The power generation amount prediction device according to any one of claims 1 to 3, wherein the power generation amount prediction unit performs the power generation amount prediction by further using the sunshine correction coefficient. The power generation amount prediction unit samples the weather condition value based on the reference weather data, corrects the sampled weather condition value according to the sunshine correction coefficient, and corrects the weather condition value based on the system coefficient. The power generation amount prediction apparatus according to claim 4, wherein power generation amount prediction is performed. A solar position calculator for calculating the position of the sun from the date and time;
The power generation amount prediction device according to claim 4, wherein the sunshine correction coefficient is associated with a position of the sun. A power generation result data selection unit for selecting the power generation result data based on the weight according to the value of the sunshine correction coefficient from the power generation result data in the power generation result data storage unit;
The power generation amount prediction apparatus according to any one of claims 4 to 6, wherein the parameter learning unit updates the system coefficient using the power generation result data selected by the power generation result data selection unit. The power generation amount prediction apparatus according to any one of claims 4 to 7, wherein the parameter learning unit updates the sunshine correction coefficient by dividing the estimated weather situation value by the corrected weather situation value. The power generation amount prediction apparatus according to any one of claims 1 to 8, wherein the weather condition value includes a sunshine value and an air temperature. The power generation amount prediction apparatus according to any one of claims 1 to 9, wherein the weather situation space correction unit performs the correction processing based on a simple average, a spatial moving average, a spatial median, or a Markov random field. Reading data from a reference weather data storage that stores reference weather data including weather situation values collected at a reference weather site;
Power generation result data storage for storing the power generation result data for each PV system, including the current value and voltage value of the power generated by the photovoltaic power generation (PV: Photovoltaic) system installed at a plurality of sites different from the reference weather site Reading data from the unit;
For each of the PV systems, reading data from a PV system coefficient storage unit that stores a system coefficient that is a coefficient for predicting a power generation amount from a weather condition value;
For each of the PV systems, a weather condition estimation step for obtaining an estimated weather condition value by estimating the weather condition at the site from the system coefficient and the power generation result data;
A weather condition space correction step for obtaining a corrected weather condition value by correcting the estimated weather condition value for each PV system according to the installation position of each PV system;
For each PV system, a parameter learning step for updating the system coefficient based on the corrected weather condition value and the power generation result data;
A power generation amount prediction method in which a computer executes a power generation prediction step of performing power generation amount prediction based on the reference weather data and the system coefficient for each PV system.

Images